Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Qwen3.6-35B-A3B: Agentic coding power, now open to all

qwen.ai

Post reply on HN
Print view View original

521–530 of 563 posts

Re: Qwen3.6-35B-A3B: Agentic coding power, now open to all

#521

Post by Der_Einzige »

reissbaker wrote: post #496
Der_Einzige wrote: post #172

Earlier quoted context omitted.

I've been telling analysts/investors for a long time that dense architectures aren't "worse" than sparse MoEs and to continue to anticipate the see-saw of releases on those two sub-architectures. Glad to continuously be vindicated on this one. For those who don't believe me. Go take a look at the logprobs of a MoE model and a dense model and let me know if you can notice anything. Researchers sure did.

Dense is (much) worse in terms of training budget. At inference time, dense is somewhat more intelligent per bit of VRAM, but much slower, so for a given compute budget it's still usually worse in terms of intelligence-per-dollar even ignoring training cost. If you're willing to spend more you're typically better off training and running a larger sparse model rather than training and running a dense one. Dense is nic…

You want to take bets on this? I'm willing to bet 500USD that an open access dense model of at least 300B is released by some lab within 3 years.

Re: Qwen3.6-35B-A3B: Agentic coding power, now open to all

#522

Post by namibj »

Glemllksdf wrote: post #203
Aurornis wrote: post #173

Earlier quoted context omitted.

Additional VRAM is needed for context. This model is a MoE model with only 3B active parameters per expert which works well with partial CPU offload. So in practice you can run the -A(N)B models on systems that have a little less VRAM than you need. The more you offload to the CPU the slower it becomes though.

Isn't that some kind of gambling if you offload random experts onto the CPU? Or is it only layers but that would affect all Experts?

Speculative decoding is already gambling.

Re: Qwen3.6-35B-A3B: Agentic coding power, now open to all

#523

Post by ElectricalUnion »

agentifysh wrote: post #411
3836293648 wrote: post #374

Qwen3.6 and Gemma4 have the same issue of never getting to the point and just getting stuck in never ending repeating thought loops. Qwen3.5 is still the best local model that works.

I think the hype around Qwen and even Gemma4 often floated for views/attention glosses over that these models have clear gaps behind what closed models offer. In short, it has its uses but it would/should not be the main driver. Will it get better, I'm sure of it, but there is too much hype and exaggeration over open source models, for one the hardware simply isn't enough at a price point where we can run something t…

I say "if we got $CURRENT_MODEL that can run under local hardware" claims are postproning BS.

What is gonna happen when that happens? They are gonna cry they need GPT-$CURRENT capabilities locally.

Now we have local models that are way better that GPT-2 (careful, this one is way too dangerous for release!) GPT3.5, in some ways better that 4, and can run on reasonably modest hardware.

Re: Qwen3.6-35B-A3B: Agentic coding power, now open to all

#524

Post by 060880 »

culi wrote: post #326
simonw wrote: post #270

I've been running this on my laptop with the Unsloth 20.9GB GGUF in LM Studio: https://huggingface.co/unsloth/Qwen3.6-35B-A3B-GGUF/blob/mai... It drew a better pelican riding a bicycle than Opus 4.7 did! https://simonwillison.net/2026/Apr/16/qwen-beats-opus/

the more I look at these images the more convinced I become that world models are the major missing piece and that these really are ultimately just stochastic sentence machines. Maybe Chomsky was right

The Chomsky argument feels like it's moving in a different direction than what's actually useful to know. Whether or not these models have "real" understanding, they're clearly capable of solving problems that were previously considered to require understanding. The more interesting question is whether world models, if they existed, would actually improve the failure modes people care about — like hallucination and planning — or whether we'd just get better stochastic sentence machines with an extra layer of abstraction on top.

Re: Qwen3.6-35B-A3B: Agentic coding power, now open to all

#525

Post by altruios »

I have moved through the local models at this size.

This one is by far the most capable. I've tried various versions of gemma4.26b, various versions of qwen3.5-27/35b (qwopus's galor),nemotron,phi,glm4.7.

This one is noticeably better as an agent. It's really good at breaking down tasks into small actionable steps, and - where there is ambiguity - asks for clarification. It's reasoning seems more solid than gemma4, tool use, multi-messaging/longer chain thinking.

I am excited to see what other versions of this model people train!

Re: Qwen3.6-35B-A3B: Agentic coding power, now open to all

#526

Post by fragmede »

johanvts wrote: post #469
bertili wrote: post #285

Earlier quoted context omitted.

It's fascinating that a $999 Mac Mini (M4 32GB) with almost similar wattage as a human brain gets us this far.

Interesting thought, I looked it up out of curiosity and fund 155w max (but realistically more like 80w sustained) for the mac under load, and just around 20watts for the brain, surprisingly almost constant whether “under load” or not.

I clocked my M4 at 108 Watts while running inference using Qwen3.6-35b-a3b via Al dente.

Re: Qwen3.6-35B-A3B: Agentic coding power, now open to all

#527

Post by nxtfari »

alecco wrote: post #189

Related interesting find on Qwen. "Qwen's base models live in a very exam-heavy basin - distinct from other base models like llama/gemma. Shown below are the embeddings from randomly sampled rollouts from ambiguous initial words like "The" and "A":" https://xcancel.com/N8Programs/status/2044408755790508113

This makes a lot of my experience with Qwen make sense. I’ve watched all the benchmarks imply how close it should be to various GPT or Claude releases, but in my own use chatting with it or trying to get it do agentic tasks it was nowhere near as smart as even GPT-3.5 for example. Meanwhile Gemma 4 casually dropped and even the 4B models were performing better than Qwen 3.5 MOE in my chats. Benchmaxxing.

Re: Qwen3.6-35B-A3B: Agentic coding power, now open to all

#528

Post by nxtfari »

3836293648 wrote: post #374

Qwen3.6 and Gemma4 have the same issue of never getting to the point and just getting stuck in never ending repeating thought loops. Qwen3.5 is still the best local model that works.

I had issues with Qwen thinking endlessly when I didn’t know I wasn’t using the temp/top_k/min_p/etc settings specified in the readme. I’ve never had an issue with Gemma 4 thinking endlessly but could possibly be the same.

Re: Qwen3.6-35B-A3B: Agentic coding power, now open to all

#529

Post by aruametello »

tjwebbnorfolk wrote: post #48
zoobab wrote: post #31

"open source" give me the training data?

The training data is the entire internet. How do you propose they ship that to you

to be fair there are some degree of "hand curation" of the data so while "it is the internet", the actual trained data is a derivation of that.

in a mild but productive analogy:

I could actually hand a K&R book C programming book + lots of specs to say "this is the linux source code" (the raw data that were all observations were made, aka "the internet") ...or just send them the "kernel the source code" (the refined training data, after a LOT of manual stuff) ... that your compiler consumes to generate the kernel. (the Open Weights model, what they actually shared)

Mildly related rant: honestly its a bit shit to say "open source model" in a "open weights" model, its like saying World of Warcraft is opensource because they gave you an executable of the game. (you can still change it, but in more restricted ways)

Re: Qwen3.6-35B-A3B: Agentic coding power, now open to all

#530

Post by onlyrealcuzzo »

altruios wrote: post #525

I have moved through the local models at this size. This one is by far the most capable. I've tried various versions of gemma4.26b, various versions of qwen3.5-27/35b (qwopus's galor),nemotron,phi,glm4.7. This one is noticeably better as an agent. It's really good at breaking down tasks into small actionable steps, and - where there is ambiguity - asks for clarification. It's reasoning seems more solid than gemma4, t…

How does it compare to CC Opus Max?
Post reply on HN